Fix tolerance when checking if diagonal is unitary

[chriseclectic]

Summary

Fixes #705

Fixes bug where diagonal gates are flagged as non-unitary due to being checked with too high a tolerance.

Adds extra diagonal gate tests

Details and comments

[adcorcol]

Maybe I'm doing something wrong, but I have implemented your change
for (const auto val : op.params) { if (!Linalg::almost_equal(std::abs(val), 1.0, 1e-7)) { throw std::invalid_argument("\"diagonal\" matrix is not unitary."); } }

within json_to_op_diagonal in operations.hpp and I still keep getting a "diagonal" matrix is not unitary' error. It does work for diagonals up to 2 or 3 qubits but fails for 4 qubits and beyond. I have tried toggling the tolerance a bit but still no success. Could this error be coming from somewhere else? Or maybe I need to change some other file that I did not notice?

(For example this 4-qubit diagonal

array([1. +0.j , 0.99999171-0.00407149j, 0.99999534-0.00305362j, 0.99991245-0.013232j , 0.99998337-0.00576793j, 0.99995159-0.00983931j, 0.99979279-0.02035611j, 0.9995338 -0.03053154j, 0.99981518-0.01922536j, 0.99950577-0.03143588j, 0.99975181-0.02227833j, 0.99917585-0.04059079j, 0.99798788-0.06340504j, 0.99713899-0.07558995j, 0.99695665-0.07795798j, 0.99536107-0.09620988j])

fails)

[chriseclectic]

@adcorcol For your 4-qubit example I am getting in error from the terra diagonal gate class that one of the entries does not have absolute value 1, so I think it's failing before it gets to the simulator.

Looking at the diagonal gate class it is checking with tolerance 1e-10, which is too high for Numpy. I think its worth raising an issue for that on Qiskit Terra.

[adcorcol]

Thanks, I will look into it and submit an issue.